Re: help!

[James Sherman Jr.]

I'll just basically reiterate what Jordi said, if initial guesses were easy to come by/not needed, then optimization would be a whole heck of a lot easier.

Just some ideas on how to get some initial guesses:
1) Is there some physical processes behind your model?  Do some parameters correspond to factors measurable by some other means?  (for example, population size, mean age, life expectancy)

2) Does your model become simpler at certain values of x?  In your first email you wrote poly1(x) + exp(poly2(x)), so as a first guess, maybe look at small values of x, then you could use something like polyfit to estimate the coeffcients of poly1 (well except for the constant term, since I'm assuming for small values of x, the second term is constant).  Or on the other hand, for large values of x, the exp term should dwarf the first term, so maybe use polyfit again on the natural log of measurements of values with a large x.

Just a couple ideas.

James Sherman

On 8/1/07, Jordi Gutiérrez Hermoso <address@hidden> wrote:

Please keep the conversation on the list, so that others may participate.

On 01/08/07, address@hidden <address@hidden> wrote:
>        Thank you for your reply. I have tried using S-PLUS, which is the
> programme R is a clone of, however its non linear optimisation requires
> starting values.

Just like Octave isn't Matlab, R isn't S (in fact, I've heard more
comments about R being better than S than the other way around). Try
checking it out anyways. It might help.

As for an initial point for nonlinear optimisation, that's just the
way it is. Optimisation is just hard, period. I can't think of any
nonlinear optimisation method that doesn't require an initial guess of
some sort. I'm sure they exist, but they just seem less common.

- Jordi G. H.
_______________________________________________
Help-octave mailing list
address@hidden
https://www.cae.wisc.edu/mailman/listinfo/help-octave